Forums

AUDIO SIGNAL NOISE DETECTION

Started by rickowens January 28, 2016
Hello everyone. I have to make a plugin that detects and analyzes the
noise in it, in order to classify the audio tracks depending on quality.
What is the most efficient way to do this? Any suggestions?
Thank you.
---------------------------------------
Posted through http://www.DSPRelated.com
>Hello everyone. I have to make a plugin that detects and analyzes the >noise in it, in order to classify the audio tracks depending on quality. >What is the most efficient way to do this? Any suggestions? >Thank you. >--------------------------------------- >Posted through http://www.DSPRelated.com
Good luck with that. Kind of reminds me of the frequent "Who is the best band?" flame wars in some music forums years back. Or my father walking in and saying "Turn off that god damned noise." when we were listening to albums. The best you can probably do is to decompose your signal to a sum of pure tones to measure how tonal it is. For a real good discussion on "What is quality?", I highly recommend the book "Zen and the Art of Motorcycle Maintenance". It has nothing to do with DSP though. Ced --------------------------------------- Posted through http://www.DSPRelated.com
"Cedron" <103185@DSPRelated> Wrote in message:
> >Hello everyone. I have to make a plugin that detects and analyzes the >>noise in it, in order to classify the audio tracks depending on quality. >>What is the most efficient way to do this? Any suggestions? >>Thank you. >>--------------------------------------- >>Posted through http://www.DSPRelated.com > > Good luck with that. Kind of reminds me of the frequent "Who is the best > band?" flame wars in some music forums years back. > > Or my father walking in and saying "Turn off that god damned noise." when > we were listening to albums. > > The best you can probably do is to decompose your signal to a sum of pure > tones to measure how tonal it is.
Suggested classification scheme: A) Garbage B) Not Garbage -- ----Android NewsGroup Reader---- http://usenet.sinaapp.com/
On Friday, January 29, 2016 at 1:19:11 AM UTC+13, rickowens wrote:
> Hello everyone. I have to make a plugin that detects and analyzes the > noise in it, in order to classify the audio tracks depending on quality. > What is the most efficient way to do this? Any suggestions? > Thank you. > --------------------------------------- > Posted through http://www.DSPRelated.com
Unless you have a period of no signal ie noise alone you will never capture the noise power.
>On Friday, January 29, 2016 at 1:19:11 AM UTC+13, rickowens wrote: >> Hello everyone. I have to make a plugin that detects and analyzes the >> noise in it, in order to classify the audio tracks depending on
quality.
>> What is the most efficient way to do this? Any suggestions? >> Thank you. >> --------------------------------------- >> Posted through http://www.DSPRelated.com > >Unless you have a period of no signal ie noise alone you will never
capture
>the noise power.
What about the Yule-Walker estimation of the autoregressive models? Do you think that the output variance is good to quantify the background broadband noise? I have to do a plugin which quantify the broadband noise of a wav track. Any suggestions? --------------------------------------- Posted through http://www.DSPRelated.com
On Friday, February 26, 2016 at 4:18:04 AM UTC+13, rickowens wrote:
> >On Friday, January 29, 2016 at 1:19:11 AM UTC+13, rickowens wrote: > >> Hello everyone. I have to make a plugin that detects and analyzes the > >> noise in it, in order to classify the audio tracks depending on > quality. > >> What is the most efficient way to do this? Any suggestions? > >> Thank you. > >> --------------------------------------- > >> Posted through http://www.DSPRelated.com > > > >Unless you have a period of no signal ie noise alone you will never > capture > >the noise power. > > What about the Yule-Walker estimation of the autoregressive models? Do you > think that the output variance is good to quantify the background > broadband noise? > > I have to do a plugin which quantify the broadband noise of a wav track. > Any suggestions? > --------------------------------------- > Posted through http://www.DSPRelated.com
You're going to need a classification algorithm and features. Google classification of just about anything. Classifiers, features,pattern regognition
I would use the STFT (windowed overlapping 1k fft blocks) and for each bin find the minimum magnitude over the entire song. This should give you an estimate of the noise, assuming it's stationary. Note this assumes the music is dynamic enough that for a particular bin there is at least one frame where the bin magnitude is dominated by the underlying stationary noise. This might preclude heavy metal music that is basically white noise for the entire duration of the song. 

Bob
I would do a Walsh Hadamard transform, find the lowest magnitude output at each time and average those lowest values over the whole audio track. Or the same thing with the FFT.  With noise all of the output magnitudes must be greater than zero, without noise some of the outputs may have zero magnitude.